Deep learning-driven Cell-to-Vehicle transfer of differential voltage and incremental capacity features for electric vehicle battery state-of-health diagnostics

Achieving Sustainable Development Goal 7 (SDG 7) requires energy-efficient electrified mobility supported by reliable and long-lived electric vehicle (EV) battery systems. However, existing battery diagnostic approaches face three persistent challenges: (i) limited transferability from laboratory cell data to real-world vehicle battery packs, (ii) reduced reliability of classical electrochemical diagnostics such as Incremental Capacity Analysis (ICA) and Differential Voltage Analysis (DVA) under fast charging, partial charging, and noisy operating conditions, and (iii) a lack of interpretable deep learning models that generalize across domains with limited labeled vehicle data. To address these challenges, this paper proposes a deep learning–driven Cell-to-Vehicle (C2V) transfer learning framework that integrates electrochemically grounded ICA/DVA representations with a hybrid CNN–LSTM–Vision Transformer (ViT) architecture. The proposed model jointly captures local electrochemical feature morphology, long-term degradation dynamics, and global voltage–capacity dependencies, enabling robust health diagnostics under realistic EV operating conditions. A domain-adaptive transfer mechanism aligns cell-level and vehicle-level feature distributions, reducing labeled data requirements by approximately 70%. Experimental validation on the NASA cell dataset and real-world EV fleet data demonstrates strong cross-domain performance, achieving RMSE values of 0.012 and 0.019 and R 2 scores of 0.987 and 0.978 at the cell and vehicle levels, respectively, corresponding to a 45%–55% error reduction compared to existing deep learning baselines. Beyond predictive accuracy, the proposed framework enables predictive maintenance, improves charging efficiency, and supports lifetime-aware battery management, with projected impacts including a 35% extension in battery cycle life and a 10% reduction in lifecycle CO 2 emissions. These results position electrochemically informed C2V diagnostics as a key enabler for sustainable, affordable, and reliable electric mobility aligned with SDG 7.